US 20060268731 A1 Abstract Methods and apparatus are provided for measuring and evaluating access link performance in IP networks that reduce the amount of required test traffic. Performance values supported by access links in a packet network that interconnects regions of an enterprise network are evaluated by (i) obtaining a plurality of test measurements for a set of N access links; (ii) formulating a matrix equation (Ax=D), where A is an invertible matrix, such as a Hadamard matrix, x is a vector of unknown access link performance parameters and D is a vector based on the plurality of test measurements; (iii) assigning one or more numeric values to one or more unproducible equations in the matrix equation based on one or more system properties; and (iv) obtaining a performance parameter value on each directed edge for the set by applying an inverse matrix A
^{−1 }to each side of of the matrix equation. Claims(20) 1. A method for evaluating performance values supported by access links in a packet network that interconnects regions of an enterprise network, said method comprising:
obtaining a plurality of test measurements for a set of N access links; formulating a matrix equation (Ax=D), where A is an invertible matrix, x is a vector of unknown access link performance parameters and D is a vector based on said plurality of test measurements; assigning one or more numeric values to one or more unproducible equations in said matrix equation based on one or more system properties; and obtaining a performance parameter value on each directed edge for the set by applying an inverse matrix A ^{−1 }to each side of Ax=D. 2. The method of 3. The method of 4. The method of 5. The method of 6. The method of 7. The method of 8. The method of 9. The method of 10. A system for evaluating performance values supported by access links in a packet network that interconnects regions of an enterprise network, comprising:
a memory; and at least one processor, coupled to the memory, operative to: obtain a plurality of test measurements for a set of N access links; formulate a matrix equation (Ax=D), where A is an invertible matrix, x is a vector of unknown access link performance parameters and D is a vector based on said plurality of test measurements; assign one or more numeric values to one or more unproducible equations in said matrix equation based on one or more system properties; and obtain a performance parameter value on each directed edge for the set by applying an inverse matrix A ^{−1 }to each side of Ax=D. 11. The system of 12. The system of 13. The system of 14. The system of 15. The system of 16. The system of 17. The system of 18. An article of manufacture for evaluating performance values supported by access links in a packet network that interconnects regions of an enterprise network, comprising a machine readable medium containing one or more programs which when executed implement the steps of:
obtaining a plurality of test measurements for a set of N access links; formulating a matrix equation (Ax=D), where A is an invertible matrix, x is a vector of unknown access link performance parameters and D is a vector based on said plurality of test measurements; assigning one or more numeric values to one or more unproducible equations in said matrix equation based on one or more system properties; and obtaining a performance parameter value on each directed edge for the set by applying an inverse matrix A ^{−1 }to each side of Ax=D. 19. The article of manufacture of 20. The article of manufacture of Description The present invention relates generally to communication methods and systems, and more particularly, to methods and systems for performance measurements supporting higher-level functions, such as admission control, path switching, performance monitoring, and problem isolation, in a packet-switched network. Generally, the Internet Protocol (IP) and IP networks have been designed to support a single, best-effort class of service. IP networks have successfully transported TCP-mediated data traffic more cost-effectively and flexibly than other popular network types, e.g., circuit-switched networks. As a result, there is a convergence effort to migrate all networked applications, such as voice and videoconferencing applications, to use IP networks as the common transport medium. Best-effort service, however, is not sufficient to meet the Quality-of-Service (QoS) needs of some of these migratory applications, especially in an enterprise environment. Hence, the IP networking industry has been developing QoS solutions with differentiated services that provide different levels of transport performance in accordance with the needs of higher level applications. A part of many QoS solutions is performance measurement, and specifically the measurement of the current delay, jitter (delay variance), and packet loss probability values for packet traffic traversing communications links connecting routers and switches and traversing multi-hop switched and routed paths through the network. Measured delay, jitter, and packet loss probability values may be used in the decision processes of higher-level functions such as an admission control system, which governs which traffic enters the network. Admission control is required, for example, to support QoS for many inelastic applications. A number of techniques have been proposed or suggested for admission control in packet-switched networks. For example, U.S. patent application Ser. No. 11/111,464, entitled “Method and Apparatus for Quality-of-Service-Based Admission Control,” incorporated by reference herein, discloses QoS-based admission control (QBAC) techniques for a packet network. Performance measurement techniques may be categorized as passive, active, or hybrid. Passive techniques measure statistics of actual bearer (non-synthetic) traffic. Active techniques probe the network with synthetic traffic. Hybrid techniques combine passive and active methods to improve accuracy and reduce active traffic loads. Probing a network with test traffic can potentially affect the performance of bearer traffic. Thus, it is important to minimize the complexity of the test architecture, i.e., to use the minimum amount of test traffic necessary to collect the required information. When it is possible to observe the traffic that enters and exits both ends of an individual link, then measuring the transport performance supported by the link is relatively straightforward. Often, however, geographically distributed enterprises contract with service providers to provide wide-area network (WAN) interconnectivity for connecting their local-area networks (LANs) together to form internetworks. In such cases, the enterprise cannot observe the traffic entering and exiting the access link termination point in the service providers' networks. Because access links are often the bottleneck links in a network and therefore primary contributors to degradation in transport QoS, performance measurements for these access links are critical for QoS management and support of higher-level functions. In this case, naive methods exist to collect the performance measurements needed to support higher level functions, but they generate too much test traffic, i.e., the test traffic volume is large enough to significantly affect the transport performance of application traffic. A need therefore exists for methods and apparatus for measuring and evaluating access link performance in IP networks that reduce the amount of required test traffic. Generally, methods and apparatus are provided for measuring and evaluating access link performance in IP networks that reduce the amount of required test traffic. According to one aspect of the invention, performance values supported by access links in a packet network that interconnects regions of an enterprise network are evaluated by (i) obtaining a plurality of test measurements for a set of N access links (e.g., at least 3N test measurements); (ii) formulating a matrix equation (Ax=D), where A is an invertible matrix, such as a Hadamard matrix, x is a vector of unknown access link performance parameters and D is a vector based on the plurality of test measurements; (iii) assigning one or more numeric values to one or more unproducible equations in the matrix equation based on one or more system properties; and (iv) obtaining a performance parameter value on each directed edge for the set by applying an inverse matrix A According to another aspect of the invention, numeric values are assigned to one or more unproducible equations in the matrix based on one or more system properties, such as a system property that recognizes that net traffic into and out of the set of access links has an expected value, such as an expected value of approximately zero. The performance values may comprise, for example, one or more of current delay, delay variance, and packet loss probability values for packet traffic traversing one or more of the access links. A more complete understanding of the present invention, as well as further features and advantages of the present invention, will be obtained by reference to the following detailed description and drawings. A typical enterprise network consists of a collection of local-area networks (LANs) interconnected using private-line WAN connectivity or service provider network connectivity, or a combination of the foregoing. Typically, the physical bandwidth of the LANs is much greater than the bandwidth of the private lines and/or the access links connecting LANS with the service provider network. Service provider networks also typically have bandwidth that is much greater than the access links. Hence, private lines and access links are usually the “bottleneck” links in a wide-area network (WAN). Bottleneck links often become congested, i.e., the current packet traffic load exceeds the link capacity. During congestion episodes, the router or switch terminating the bottleneck link enqueues incoming packets. Packet queueing causes additional delays beyond the “speed-of-light” propagation delays and also induces jitter in individual microflows. During prolonged episodes of congestion, packet queues may fill up, in which case incoming packets are dropped. Queueing delay, jitter, and packet loss sourced by congested bottleneck links is a major source of QoS degradation for some applications, such as IP telephony using the Voice-over-IP (VoIP) protocol, IP-based videoconferencing, and so-called “IP circuit emulation” applications, such as FAX-over-IP (FoIP) and Teletype-over-IP (TTYoIP). Thus, a critical component of any enterprise's QoS management solution is timely measurement and estimation of the current delay, jitter, and loss probability values sourced by bottleneck links. When private-line interconnectivity is used, the enterprise often owns, controls, or otherwise has unfettered access to the routers, switches, and/or host devices that terminate both ends of the private line. In this case, it is straightforward for the enterprise to instrument both ends of the private line and then measure the delay, jitter, and loss probability experienced by actual packet flows or synthetic packet traffic traversing the link. The measured values may then be used as input to higher-level QoS management functions such as admission control, performance monitoring, problem isolation, and path switching. For cost and availability reasons, enterprises may use so-called service provider interconnectivity instead of private lines to interconnect their LANs. A service provider owns a packet-switching network infrastructure and offers packet transport services to clients, such as geographically distributed enterprises. Enterprises purchase or lease access links that connect their LANs to the closest access point, or point-of-presence (POP), of the service provider network. Packet traffic that is sourced and sunk in two different LANs traverses an access link from the source LAN to the service provider network, and upon exit from the service provider network traverse another access link to reach the destination LAN. As the access links are bottleneck links, a QoS management solution will want to know the current transport performance (delay, jitter, loss probability) supported by the access links. Unlike the case of private-line interconnectivity, enterprises neither own nor have access to the switching devices that terminate the access links at the service provider POPs, so the enterprise cannot instrument both ends of an access link and thereby directly measure the link's transport performance attributes. The enterprise can instrument the termination points of access links that terminate at devices that they own. Hence, the enterprise can measure the delay, jitter, and loss probability values for the path between two LANs. Paths between LANs are composed of two access links and links internal to the service provider network. Given values of delay, jitter, and loss probability for a single path, it is not possible to know the individual contributions of each access link to the path's performance values. For example, suppose the measured delay across a path is 100 ms. The delay contributed by the two access links is represented as the variables x One method for collecting necessary performance measurements to support some higher-level applications is to measure the performance of each path between every pair of LANs in the enterprise network. Such a method obviates the need for performance information on individual links. The problem with this approach is that it does not scale, as follows: If there are N LANs in the WAN, and therefore N access links (there may be more in a multi-homed configuration), then there is on the order of N Another class of methods for estimating the contributions of individual links are referred to as tomographic methods. The method of the present invention is a tomographic method, but it differs significantly from existing tomographic methods in scope and in technique. Existing tomographic methods estimate the performance of every internal link in the target network, compute network topology, use complex statistical inference techniques to estimate performance parameters, and are not suitable for providing accurate and timely performance information for higher-level functions, such as admission control. An example of an existing tomographic method is found in United States Patent Application Serial No. 20020116154 entitled “Network Tomography Using Closed Spaced Unicast Packets.” For a typical enterprise applying QoS management, neither performance information on every internal link in the network nor network topology information is necessary for QoS management. An enterprise has no control over the links in the service provider network, nor can the enterprise dictate the service provider topology. Thus, the enterprise cannot take action to mitigate a performance problem. The most important links for an enterprise to measure are the access links. An existing tomographic method will provide performance information on access links but will do so using more traffic and computational processing resources than is necessary, and may not be able to provide timely information without using a large amount of test traffic. The present invention uses a minimal amount of test traffic and a simple, non-statistical technique to compute access link performance information, and is therefore well-suited to support higher-level QoS management functions such as admission control. More formally, the links of interest in a network (the bottleneck links) and the collection of measurement tests used to gather performance data for all of the links of interest are said to form a test architecture. The complexity of a test architecture measures the number of tests as a function of the number of links of interest. For example, the test architecture described above that measures every path between each pair of LANs contains N access links and has a complexity of O(N As described below, the present invention includes a method for estimating performance values of the access links in a WAN using a test architecture with complexity function that is linear in the number of access links. Specifically, if there are A access links connecting to service provider networks, then the number of necessary test measurements is of the same order as the number of access links, or equivalently the complexity of the corresponding test architecture is O(A). Note that A may exceed N, the number of LANs, because enterprises often use multihoming, i.e., there are multiple access links per LAN because the enterprise contracts with two or more service providers. Efficient Test Architectures for Networks using Service Provider Interconnectivity The present invention affords an efficient test architecture, i.e., a test architecture with O(A) complexity, that is applicable to networks using service provider networks for interconnectivity between LANs. The disclosed test architecture for the case of service provider connectivity reduces the test architecture complexity to O(A). Without loss of generality, it is assumed that the access links are bidirectional. Since the ingress and egress traffic traversing each access link are not identical, however, an access link is modeled as a pair of unidirectional links x In one exemplary embodiment of the present tomographic method, four access links are randomly selected, and four nodes A The intent of the tests T It is noted that this system has eight equations and eight unknowns (the x The equation Ax=D in x has the solution x=A Typically, such situations are resolved by formulating a linearly independent equation in the unknowns based on one or more system properties and replacing one of the rows of A with a row corresponding to the independent equation. For example, state probability systems with unknowns p The method of the present invention first identifies a matrix that is known to be invertible and then map tests T Consider the Hadamard matrices H Hadamard matrices are non-singular; in fact, H Applying H Every row of H Also note that Equation (3) uses delay measurements D For this exemplary Hadamard matrix and test set, Equation 4 is the independent (unproducible) equation that must be assigned a numeric value by identifying it with some system property, which will then admit a solution for the unknowns x. Assume that the distribution of traffic load is random, that the delay values x for the (arbitrarily selected) access links are uncorrelated, and for simplicity that the access links all have the same capacity. Partition the set of unknowns {x The example above is for the case where the number of access links A=4, but the method generalizes to A=2 The above exemplary four-access-link solution provides a solution for the entire enterprise network using service provider interconnectivity As shown in 3N tests, such as tests T A matrix A that is known to be invertible is identified during step In step In step To estimate the end-to-end delay for any path traversing the service provider network, under the assumption that the majority of the delay due to resource contention is at the access links, and with known or upper bound values of delays across the service provider network core (typically available as part of Service Level Agreements (SLAs)), add the delay values of the two access links along the path to the core delay and any other known path delays to get the estimated end-to-end delay. The test architecture associated with this solution has complexity O(A), where A is the total number of access links to the provider network. This is readily apparent, as each set of four (bidirectional) access links requires 12 tests, for a total of 3A tests for the entire set of access links. Larger Hadamard matrices (i.e., H For simplicity, the system condition assumptions used in Eq. 5 include an assumption that the access link capacities are equal. If the access link capacities are not equal, then appropriate adaptations may be made to the process. One adaptation would be to partition the set of access links by link capacity and then apply the process to each partition, in which case the assumption of equal link capacities holds. Alternatively, the measured delay value D If A mod 4≢0, or more generally, if A mod 2 Another application of this approach is in scenarios where round-trip testing is used, i.e., performance statistics are gathered at the source point only and not the sink (the sink functions only as a volley/loopback mechanism). Only the round-trip delay is measured, but there may be a need to distinguish between the one-way delays. An appropriate modification to the disclosed method can determine one-way delays from round-trip measurement data, as would be apparent to a person of ordinary skill in the art. System and Article of Manufacture Details As is known in the art, the methods and apparatus discussed herein may be distributed as an article of manufacture that itself comprises a computer readable medium having computer readable code means embodied thereon. The computer readable program code means is operable, in conjunction with a computer system, to carry out all or some of the steps to perform the methods or create the apparatuses discussed herein. The computer readable medium may be a recordable medium (e.g., floppy disks, hard drives, compact disks, or memory cards) or may be a transmission medium (e.g., a network comprising fiber-optics, the world-wide web, cables, or a wireless channel using time-division multiple access, code-division multiple access, or other radio-frequency channel). Any medium known or developed that can store information suitable for use with a computer system may be used. The computer-readable code means is any mechanism for allowing a computer to read instructions and data, such as magnetic variations on a magnetic media or height variations on the surface of a compact disk. The computer systems and servers described herein each contain a memory that will configure associated processors to implement the methods, steps, and functions disclosed herein. The memories could be distributed or local and the processors could be distributed or singular. The memories could be implemented as an electrical, magnetic or optical memory, or any combination of these or other types of storage devices. Moreover, the term “memory” should be construed broadly enough to encompass any information able to be read from or written to an address in the addressable space accessed by an associated processor. With this definition, information on a network is still within a memory because the associated processor can retrieve the information from the network. It is to be understood that the embodiments and variations shown and described herein are merely illustrative of the principles of this invention and that various modifications may be implemented by those skilled in the art without departing from the scope and spirit of the invention. Referenced by
Classifications
Legal Events
Rotate |